Optimization of Event Reconstruction within the TICL Framework in CMSSW at the CMS experiment

The CMS experiment will use the TICL framework for the High Granular- ity Calorimeter (HGCAL) reconstruction and the linking with tracks in the High- Luminosity Phase of the LHC. The current implementation is CPU- based, but the goal is to employ hardware accelerators such as NVIDIA and AMD GPUs to en- hance the performance. In this project, we worked on benchmarking the TICL framework using profiling tools and on porting part of the reconstruction to hetero- geneous hardware. Specifically, we wrote heterogeneous code for the CLUE (Clus- tering of Energy) algorithm. CLUE clusters energy deposits inside HGCAL in 2D (one layer of the calorimeter) or in 3D (all the layers). We profiled the HGCAL re- construction and moved the input of the CLUE3D algorithm to heterogeneous code using alpaka.